ADE20K-Seq
收藏资源简介:
ADE20K-Seq数据集是由ADE20K数据集扩展而来,专门用于评估序列图像的交互式分割任务。该数据集包含七个不同类别的任务,每个类别包含超过100张图像,涵盖了从大型物体到小型物体的广泛类别。数据集的创建旨在解决现有数据集在处理序列图像时的不足,特别是缺乏对序列信息的利用。通过标注任务级别的点击和掩码信息,ADE20K-Seq数据集为模型提供了丰富的上下文信息,适用于评估模型在交互式图像分割中的表现,尤其是在处理连续图像序列时的准确性和效率。
ADE20K-Seq is an extension of the ADE20K dataset, specifically developed for evaluating interactive segmentation tasks on sequential images. This dataset includes tasks across seven distinct categories, with each category containing over 100 images, covering a wide range of object scales from large to small items. The dataset was created to address the limitations of existing datasets when dealing with sequential images, particularly the lack of exploitation of sequential information. By featuring task-level click and mask annotations, the ADE20K-Seq dataset provides rich contextual information for models, making it ideal for assessing model performance in interactive image segmentation, especially the accuracy and efficiency when processing continuous image sequences.

- 1SPT: Sequence Prompt Transformer for Interactive Image Segmentation中国科学院计算技术研究所 · 2024年



